List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yashi Li, Huihui Ye, Huafeng Liu
Abstract
Accurate localization of the ectopic pacing is the key to effective catheter ablation for curing cardiac diseases such as premature ventricular contraction (PVC) and tachycardia. Invasive localization method can achieve high precision but has disadvantages of high risk, high cost and its time-consuming process, therefore a non-invasive and convenient localization method is in demand. Noninvasive methods have been developed to utilize electrophysiological information provided by 12-lead electrocardiogram (ECG), and most of them are purely based on end-to-end data-driven architecture. This architecture generally needs a large and comprehensive labeled dataset, which is very difficult to obtain for whole ventricular ectopic beats in clinical setting. To address this issue, we propose a framework that combines cardiac forward-solution simulation and deep learning network for patient-specific noninvasive ectopic pacing localization. For each patient, it only requires his/her own CT images to establish specific heart-torso model and to simulate various ECG data from different ectopic pacing locations, and uses this simulated ECG data as training dataset for our designed network. The network mainly contains time-frequency fusion module and local-global feature extraction module. Five PVC patient ECG data are tested with high precision and accuracy for ectopic pacing localization, which shows its high-potential in clinical setting.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43990-2_19
SharedIt: https://rdcu.be/dnwLu
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a framework that combines cardiac forward-solution simulation and deep learning network for patient-specific noninvasive ectopic pacing localization to effectively treat cardiac diseases. The framework only requires the patient’s own CT images to establish a specific heart-torso model and to simulate various ECG data from different ectopic pacing locations. The proposed method is tested with five patient ECG data and demonstrates high precision and accuracy for ectopic pacing localization, indicating high potential in clinical applications.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- very clever idea
- The framework offers a non-invasive and convenient localization method
- The framework establishes a specific heart-torso model for each patient, providing personalized and patient-specific localization
- The framework uses a simulation-based approach to generate ECG data, making it possible to create a comprehensive labeled dataset for training the deep learning network in a cheap way
- Unlike most end-to-end data-driven architectures, the proposed framework requires a relatively smaller and specific dataset, making it easier to obtain in a clinical setting
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
- too few subject to be able to conclude something about the robustness of the results. Moreover, the method in fact relies on a specific labeled dataset generated through simulation, which is not fully represent the diversity of real-world ECG data.
- The deep learning network used in the framework is a black box, and the process of how the network arrives at its predictions is not easily interpretable by clinicians, potentially limiting its adoption in clinical practice.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
limited
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html
The proposed framework is tested on a small dataset of only five PVC patients, and its performance on a larger and more diverse dataset is yet to be determined. Some effort could be devised in this direction
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I balanced more the pros than the cons. Some little effort in the explainability part could improve this recommendation
- Reviewer confidence
Confident but not absolutely certain
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #2
- Please describe the contribution of the paper
The authors propose a novel deep-learning-based method to non-invasively locate cardiac ectopic pacing sites using 12-lead ECGs. The system builds upon time and frequency information as well as a local & gobal feature extractor. The method is solely trained on synthetic data and is evaluated on 5 clinical PVC patients.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The method is interesting and the incorporation of time + frequency information as well as local + global information is well motivated and of particular merit for the task of ectopic pacing localization.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
There are several weaknesses that should be addressed by the authors:
- There are some grammar and spelling errors
- The graphics are too small and could be of higher quality
- Essential information about the training data generation and the network are missing
- A discussion of the results is missing
- The method is only evaluated on a small population of real patients
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
The paper is missing curcial information to replicate the results (see detailed comments).
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html
While the proposed method seems to hold merits for the task at hand, the manuscript currently lacks information required to fully assess the method’s performance. I would kindly ask the authors to address the following questions and concerns:
- I would kindly ask the authors to run a spell & grammar checker over the manuscript
- Please also revise the graphics so that they are larger in size and easy to read
- p.3: Could the authors please clarify which forward solution has been used in the experiments? If only one of the two models is considered, I would suggest to remove the other one and use the new space to increase the graphics and strengthen the text
- Please provide all information about the neural network that are necessary to replicate the results, e.g., layer sizes, strides (for Conv1D), optimizer, learning rate, batch size, etc.
- Please also clarify the differences between the networks in the ablation experiment.
- Please provide more details on the data generation process. How many geometries were used? How were the ectopic pacing locations chosen? How were the underlying model parameters chosen?
- Please also add a discussion of the results and the limitations of the work. I would be particularly interested to get some insights into the performance gap between synthetic and real data. Are the discrepancies due to noise in the clinical data? How would the method perform under changes to the ECG electrode positions?
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
3
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Even though the topic seems of relevance and the combination of different feature extractors is well motivated, the manuscript in its current form lacks important information to completely assess the current performance. In addition, it would require some improvements of the written and visual presentation.
- Reviewer confidence
Confident but not absolutely certain
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
Authors introduce a DL framework for noninvasive prediction of ectopic pacing from 12-lead ECG data. The model consists of a feature extraction and feature fusion part, and the loss function is defined as the sum of the squared Euclidean distance between the predicted and labeled coordinates. The feature extraction module combines local and global information using CNNs and gated recurrent units (GRU) with an additive self-attention mechanism to capture long-range dependencies. The Fourier transform is also used to extract frequency domain information and fuse it with the time domain information for better accuracy in localization.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1) The paper proposes a noninvasive method for predicting the location of ectopic pacing which is a challenging and important problem. 2) The proposed framework is trained only on ECG simulation data but shows great performance on clinical data. Given that obtaining large amount of clinical data is a challenging task, this method seems promising. 3) Considering both time domain and frequency domain features for analyzing the ECG data. 4) Utilizing CNNs for capturing local features and GRUs for capturing long distance signal dependencies is interesting.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
1) The organization of methodology section is a bit confusing. A big portion of the methodology section is dedicated to explaining generation of the simulation data. The contributions listed in the introduction section and the organization of section 2 give the wrong impression that the method is a two stage framework but I believe the methodological contribution and the essence of this manuscript is explained in section 2.3 and section 2.2 could be summarized.
2) Lack of some details in the experiment section. In section 3.1, “comparison experiment result” what is the dateset used for testing and comparing different models? How is the VAENet model trained? Is it also trained on the same simulation data or only tested on this datasets?
3) Fig.3 could be revised to be more problem specific and include the input and output in the problem context to be more self explanatory.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
According to the authors responses, code and data are not going to be publicly available.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html
Please refer to the weaknesses listed. But overall, authors should include more details in the experiment section and clearly mention what dataset is used for each experiment.
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Interesting method for an important problem and promising results on clinical data. But some improvements are required to make this manuscript suitable for MICCAI which I believe some of them could be addressed in the rebuttal.
- Reviewer confidence
Confident but not absolutely certain
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Primary Meta-Review
- Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The paper proposes a framework that combines cardiac forward-solution simulation and deep learning network for patient-specific noninvasive ectopic pacing localization to effectively treat cardiac diseases. The framework only requires the patient’s own CT images to establish a specific heart-torso model and to simulate various ECG data from different ectopic pacing locations. The proposed method is tested with five patient ECG data and demonstrates high precision and accuracy for ectopic pacing localization, indicating high potential in clinical applications.
The method is interesting and the incorporation of time + frequency information as well as local + global information is well motivated and of particular merit for the task of ectopic pacing localization. Authors introduce a DL framework for noninvasive prediction of ectopic pacing from 12-lead ECG data. The model consists of a feature extraction and feature fusion part, and the loss function is defined as the sum of the squared Euclidean distance between the predicted and labeled coordinates. The feature extraction module combines local and global information using CNNs and gated recurrent units (GRU) with an additive self-attention mechanism to capture long-range dependencies. The Fourier transform is also used to extract frequency domain information and fuse it with the time domain information for better accuracy in localization.
The authors propose a novel deep-learning-based method to non-invasively locate cardiac ectopic pacing sites using 12-lead ECGs. The system builds upon time and frequency information as well as a local & gobal feature extractor. The method is solely trained on synthetic data and is evaluated on 5 clinical PVC patients.
Strengths of the paper:
- The paper proposes a noninvasive method for predicting the location of ectopic pacing which is a challenging and important problem.
- The framework offers a non-invasive and convenient localisation method
- The framework establishes a specific heart-torso model for each patient, providing personalised and patient-specific localisation
- The framework uses a simulation-based approach to generate ECG data, making it possible to create a comprehensive labeled dataset for training the deep learning network in a cheap way
- Unlike most end-to-end data-driven architectures, the proposed framework requires a relatively smaller and specific dataset, making it easier to obtain in a clinical setting
- The proposed framework is trained only on ECG simulation data but shows great performance on clinical data.
- Considering both time domain and frequency domain features for analyzing the ECG data.
- Utilising CNNs for capturing local features and GRUs for capturing long distance signal dependencies is interesting.
Weaknesses of the paper:
- It would be advise to add a paragraph or some experiments on how compare the real-world ECG data with the simulated data.
- The method is only evaluated on a small population of real patients, it makes it hard to conclude about the robustness of the results.
- Essential information about the training data generation and the network are missing
- Please add a discussion of the results
- Please increase the size of graphics are their quality
- Please double check the grammar and correct any spelling mistakes.
- Please revise the organisation of methodology section for better clarity
Recommendation: Authors proposed an interesting topic that combine different feature extractors. However, the manuscript in its current form lacks important information to completely assess the current performance. In addition, it would require some improvements of the written and visual presentation.
Author Feedback
Thanks for your careful and valuable comments, we have categorized the concerns into the following categories. 1.Dataset is limited: In our proposed framework, CT images from one patient are required to simulate various patient-specific ECG data with different ectopic pacing locations to train the network, and the real ECG data from the same patient as CT images are fed into the patient-specific network to obtain the real ectopic pacing location, and its localization accuracy is compared with golden standard location measured by invasive 3-D mapping systems. Thus due to the specific requirements, we collaborate with our local hospital to collect PVC patient data, and only 5 datasets are used for now, and hopefully more datasets will be included in the future. Although a limited number of datasets used, our proposed patient-specific method shows great accuracy as compared with the golden standard invasive localization method for all the 5 patients, and demonstrates its feasibility of accurate localization. 2.Clarification of methods and experiments: We agree the section 2.2 part is confusing since we only use the second method here, thus we will delete the introduction of the first method. Here’s the some details of our method: first, for each patient, we use CT images to get the epi-endocardium contours, and then manually traced to create triangular surface meshes while a two-level Gaussian smoothing algorithm is applied. Chose a heart node as source, solving partial differential eq.2~5 to obtain whole heart’s TMPs, then perform the same operation on all nodes. The longitudinal and transverse tensors of the diffusion tensor D are set to 4 and 1. Torso geometric model is built by the location of 120 electrodes recorded in CT scan. The 12-lead ECG signals can be extracted from the body surface potential recorded by them. A transfer matrix is computed from the geometric models using FEM in SCIRUN. Details of network in Section 2.3: Conv1D layers with kernel sizes of 5, 3, and 3; strides of 2, 2, and 1; output dimension of 32, 64, and 128 with ReLU activation. Training uses the Adam optimizer with a learning rate of 1e-4 and a batch size of 32. The datasets in Section3.1’comparison experiment’ are both generated based on the heart torso model in ECGSIM software. And models (e.g., VAENet) for comparison are based on four methods mentioned in a paper on cardiac ECG inverse imaging (ref24). The differences between different networks in Section3.2’ablation experiment’: CNN: Remove the global feature extraction module while other unchanged. Signal: The input consists only original ECGs without the feature fusion part for spectral information while other unchanged. GRU+Attention: Remove the local feature extraction module while other unchanged.
- Discussion added: Our paper proposes a non-invasive framework for PVC localization, showing promising results on clinical data. However, more datasets need to be added to verify robustness, and complex simulations including a broader range of cardiac conditions is required in the future. Mismatches between the simulation data and real data primarily stem from noise, which we demonstrate the framework’s robustness to in Fig.3. Slight variations in physiological parameters and consistent electrode placement in clinical practice can also contribute to the discrepancies. Though DL is a black box, the feature extraction is designed to transform visual features of ECGs used for clinical judgments into computer-understanding representations for decision-making. 4: Writing and graphic improvement: We have ask for help from native speakers to carefully proofread to avoid grammar and spelling errors; we have generated higher-quality images especially for Fig.4 for better visualization; we have added more details like data dimensions and each layer’s detailed parameter information for Fig.3.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
This paper proposes a framework that combines cardiac forward-solution simulation and deep learning network for patient-specific non-invasive ectopic pacing localization to effectively treat cardiac diseases. I think the rebuttal overall did a decent job of responding to most of the concerns, however, the AC still has some concern regarding the use of a small population of real patients, and the robustness of the results. Thus, the AC votes for rejecting this paper.
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
While the proposed work suffers from the lack of sample size and consequently more solid validation, the novel and interesting concept of the paper, plus the potential clinical implication (as many cardiac patients would have CT scans) makes it suitable for a MICCAI presentation.
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The authors addressed most of the reviewers comments in the rebuttal. R2 being the most critical reviewer has not updated scores after rebuttal, however, I went through the comments and find that none of the remarks are major and would speak against acceptance.The authors are encourage to improve presentation and style, as already discussed.